Build editable transcript and subtitle timelines with BaoCut
Transcribe supplied media, polish timed text, add optional translated subtitles, and export deliverables while retaining a resumable, editable video project.
npx skills add agentskillexchange/skills --skill build-editable-transcript-and-subtitle-timelines-with-baocut
Use BaoCut’s subtitle workflow when an agent must turn supplied audio or video into a reviewable transcript and editable subtitle timeline. Confirm source media, source and target languages, terminology, subtitle format, and requested exports. This entry covers transcription, correction, optional translation, caption layers and delivery for a specific video; it does not cover generic desktop editing, image generation, dubbing, motion graphics or operating a hosted video service.
What this skill actually does
Install the packaged BaoCut app from upstream releases for macOS 14 or later on Apple Silicon, or the appropriate Windows x64 build. Packaged apps contain the runtime and native workers; FFmpeg and ffprobe remain external requirements. Agent features require an installed, signed-in engine such as Claude Code or Codex CLI. Check baocut version and baocut status; stop on an interface mismatch or missing required capability. Use the documented baocut skill install command and its installed help to choose your agent. No source build is required for the packaged route.
Read the root BaoCut skill and subtitle-workflow instructions. Inspect an existing target video before making changes, or import the supplied media into a new video. Transcribe once, correct words and terminology without inventing timestamps, and only then translate and create original-language or bilingual caption layers. Check existing layers before creating another. Preserve source media, retain uncertain speaker identities, and report unresolved recognition or translation errors. Resume from current video state on later turns instead of repeating import or transcription. Adopt source chapters where available; chapter failure should not prevent the requested subtitle deliverable.
Through MCP, connect with baocut mcp install --agent codex --level ask, substituting a documented supported host as appropriate, after the operator authorizes the configuration change. MCP is a scoped external-client surface: choose an accessible project and video, import local media through BaoCut or the CLI first, and do not pass a local file directly to MCP transcription. For jobs, retain the job ID and repeat jobs_wait until terminal state; use a timeout no greater than 50 seconds. MCP export paths remain within project exports. Never bypass denied approvals or video scope.
Export SRT, WebVTT or a captioned video only as requested. Return project/video identity, caption-layer counts, export paths and remaining review needs. Invoke this workflow instead of manual desktop editing when you need a repeatable agent-led sequence that preserves the editable timeline and can resume after an interruption. Auto Subs emphasizes subtitle generation and editor handoff; Chengfeng focuses on talking-head cutting. Neither essentially replaces this runtime-backed document-to-caption sequence with project scoping and resumable state.
Multi-Framework is supported by the documented skill/CLI routes for multiple agents and optional MCP connection; MCP is not mandatory. The exact repository had 1,010 stars and 91 forks, with release baocut-v3.6.1-build.75 published October 14, 2026. BaoCut uses a source-available Community License; consult upstream terms before redistribution or service deployment. Local and remote model providers have separate requirements. No app, provider, model or export was executed during metadata verification, and no security review is claimed. Upstream subtitle workflow.